Journal of Experimental Social Psychology 46 (2010) 736–742
Contents lists available at ScienceDirect
Journal of Experimental Social Psychology
journal homepage: www.elsevier.com/locate/jesp
Framing prisoners and chickens: Valence effects in the prisoner’s dilemma
and the chicken game
Peter de Heus a,*, Niek Hoogervorst b, Eric van Dijk a
a
b
Department of Psychology, University of Leiden, The Netherlands
Department of Business-Society Management, Rotterdam School of Management, Erasmus University, The Netherlands
a r t i c l e
i n f o
Article history:
Received 15 December 2008
Revised 8 March 2010
Available online 4 May 2010
Keywords:
Social dilemmas
Prospect theory
Framing
Valence effects
a b s t r a c t
In an experimental study, we investigated how decisions in social dilemmas are affected by the valence of
outcomes that are at stake. Prospect theory states that individuals are risk-averse when outcomes are
framed as gains, and risk-seeking when outcomes are framed as losses. On the basis of this framework,
previous research on social dilemmas has addressed the question of whether people are more cooperative
in the negative domain than in the positive domain, but this research has led to inconsistent results. A
possible explanation for this is that in many social dilemmas it is unclear whether cooperation or defection is the risky choice. In the current paper, we compare the well-studied prisoner’s dilemma with the
less studied chicken game. Whereas in the prisoner’s dilemma it is unclear what constitutes the risky
option, in the chicken game the risky option is quite clear. Consistent with predictions, we found in
the chicken game more defection in the loss frame than in the gain frame, but no difference between
the gain and loss frame in the prisoner’s dilemma. Moreover, choices were predicted by risk attitude
in the chicken game, but not in the prisoner’s dilemma.
Ó 2010 Elsevier Inc. All rights reserved.
Introduction
As social beings, we frequently encounter situations where our
own interests conflict with the interests of others. Social dilemmas
are situations in which personal and collective interests are at odds
(for overviews, see e.g. Komorita & Parks, 1995; Kopelman, Weber,
& Messick, 2002; Messick & Brewer, 1983). In the current article,
we argue that how people deal with such dilemmas depends on
(a) the valence of outcomes that are at stake, and (b) the type of
dilemma people face.
Sometimes the conflict between personal and collective interests concentrates on negative outcomes. For example, in the original description of the prisoner’s dilemma, two prisoners have to
make a decision that determines for how many years they will
be sentenced. On other occasions, the outcomes may be positive,
for example, when fishermen have the choice between harvesting
more or less fish from a sea in danger of over-fishing. Are people
more cooperative when the social dilemma is about negative
rather than positive outcomes? In the present study we argue that
the effect of valence on cooperation is dependent on the structural
characteristics of the dilemma. Following prospect theory (Kahneman & Tversky, 1979, 1984) we will argue that the effect of va* Corresponding author. Address: Department of Social and Organizational
Psychology, University of Leiden, P.O. Box 9555, 2300 RB Leiden, The Netherlands.
Fax: +31 71 5273619.
E-mail address: [email protected] (P. de Heus).
0022-1031/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved.
doi:10.1016/j.jesp.2010.04.013
lence will be most pronounced when the dilemma involves a
clear choice between a risky and a non-risky decision. To demonstrate this, we compare behavior in two types of dilemmas: the
prisoner’s dilemma and the chicken game.
Prisoner’s dilemma and chicken game
In the prisoner’s dilemma, two persons have to choose independently from each other between cooperation (C) and defection (D).
If both players choose C, they both get the reward payoff (R), which
is better than the punishment payoff (P) for mutual defection, so
for both players mutual cooperation is better than mutual defection. However, in a one-sided defection the temptation payoff (T)
for the defector is even better than the mutual cooperation payoff,
while the sucker payoff (S) for the cooperator in this exchange is
even worse than the mutual defection payoff. In brief, a prisoner’s
dilemma is defined by the payoff structure T > R > P > S (Fig. 1). Because of this payoff structure, for each individual defection always
pays better than cooperation, regardless of whether the other
chooses cooperation (since T > R) or defection (since P > S). However, if both players follow their self-interest, both will be worse
off than if they both had chosen to cooperate (since P < R). Real-life
examples of prisoner’s dilemmas are two gas stations deciding
whether or not to start a price war (Murnighan, 1991), or World
War I soldiers in the trenches choosing whether or not to open
serious fire at the enemy (Axelrod, 1984).
737
P. de Heus et al. / Journal of Experimental Social Psychology 46 (2010) 736–742
Other
C
C
Gain Self
frame
Other
C
C
D
R
Other
C
D
2
3
C
0
2
1
0
3
D
1
1
3
0
0
T
S
R
Other
P
S
T
3
1
2
Self
D
Self
D
D
2
C
Other
C
D
D
P
C
Loss Self
frame
-1
0
C
-3
-1
-1
0
-2
-1
Self
D
-3
0
-2
-2
D
-2
0
-3
-3
Fig. 1. Gain-framed and loss-framed payoff structures for the prisoner’s dilemma and the chicken game.
The chicken game, also known as hawk-and-dove (Maynard Smith
& Price, 1973), resembles the prisoner’s dilemma in many respects.
Again players have to choose between cooperation and defection.
The payoff structure of the chicken game, T > R > S > P, is like the prisoner’s dilemma in the sense that the best and second-best payoff are
T (for one-sided defection) and R (for mutual cooperation) respectively. The difference is in the ranking of P and S. In chicken games
the mutual defection payoff (P) is worse than the payoff for onesided cooperation (S), so if the other defects, cooperation pays better
than defection. Many social situations have the payoff structure of a
chicken game. For example, if both parties in marital conflict choose
escalation to full conflict in order to get their way (mutual defection),
this may be very harmful to both, so trying to reach a compromise
(mutual cooperation) is usually preferable over mutual defection.
However, one-sided defection can be a very effective power tactic,
at least in the short run, if the partner prefers giving in to full conflict.
Nations using the threat of nuclear war, management and unions
heading for a strike, children doing dangerous things in order to
show their toughness, chicken games can be seen everywhere.
Although it is our impression (admittedly hard to prove) that the
chicken game is more ubiquitous in social life than the prisoner’s dilemma, it has received much less attention in psychology and other
social sciences. Searching for ‘‘prisoner’s dilemma” or ‘‘chicken
game” (or ‘‘chicken dilemma”, or ‘‘game of chicken”) in the Psychinfo
database on February 19, 2010, led to 1133 hits for the prisoner’s dilemma, against 49 for the chicken game.
The relatively small difference between the chicken game and the
prisoner’s dilemma (S > P versus P > S) leads to strongly divergent
strategic possibilities. The payoff structure of the prisoner’s dilemma, at least in the one-shot version, works strongly in favor of mutual
defection. If both you and your opponent always get more after
choosing D than after C, both fear (for the worst outcome S) and greed
(for the best outcome T) lead to defection. In the language of game
theory, defection is the dominant choice in the (one-shot) prisoner’s
dilemma, because self-interested players will always choose defection. In the chicken game, conditions are more favorable for cooperation. Greed may still lead to defection, but fear no longer does,
because in chicken the safe choice (which avoids the worst possible
outcome) is cooperation. In agreement with this analysis, higher
cooperation rates have been reported for the chicken game than
for the prisoner’s dilemma, both in two-person (Rapoport &
Chammah, 1969) and in N-person (Liebrand, Wilke, Vogel, &
Wolters, 1986; Wit & Wilke, 1992) situations.
Valence effects
So how will the valence of outcomes affect decisions in prisoner’s dilemmas and chicken games? According to prospect theory
(Kahneman & Tversky, 1979, 1984), people are risk-seeking in the
negative domain and risk-averse in the positive domain. To predict
the effects of valence on cooperation one therefore first of all needs
to identify risk-seeking and risk aversive behavior.
In this respect, we will briefly discuss previous social dilemma
research, in which prospect theory was primarily invoked to
understand differences between public good and resource dilemmas. In resource dilemmas (or take-some dilemmas), people can increase their outcomes by harvesting from a common pool, whereas
in public good dilemmas (or give-some dilemmas) they decide how
many from their own resources they contribute to a common pool
(e.g., van Dijk & Wilke, 1995). Both games share a similar conflict
between personal and collective interest. If too much is taken from,
or not enough is given to the common pool, all will be worse off
than when they had shown more restraint or generosity. The
games are different, however, in presentation of the outcome
structure. The resource dilemma is characterized by a positive
frame, the public good dilemma by a negative frame.
So what is the risky option in these dilemmas? Brewer and
Kramer (1986) reasoned that in both types of social dilemmas
defection (i.e., taking much or giving little) is the more risky choice,
because it makes the worst possible collective outcome more
likely. Subsequent theorizing, however, questioned this conclusion,
by reasoning that one could also conclude that cooperation is more
risky. For example, in a public good dilemma, contributing could be
seen as risky because one’s contributions will be wasted if the public good is not provided. Based on these considerations, some
researchers concluded that it is very difficult or even impossible
to generate predictions from prospect theory (e.g. van Dijk &
Wilke, 1995). In line with this reservation, empirical research has
shown very inconsistent findings. Whereas some studies (e.g.
Brewer & Kramer, 1986; McCusker & Carnevale, 1995) found that
participants were less cooperative in public good dilemmas than
in resource dilemmas, other studies found no difference (e.g. Rutte,
Wilke, & Messick, 1987) or even a difference in the opposite direction (e.g. Komorita & Carnevale, 1992, Experiment 3). In a large
meta-analysis on framing effects, Kühberger (1998) concluded that
game theory designs, as he called the kind of studies described
above, do not produce a framing effect at all.
Despite these objections, we will argue that it is possible to apply prospect theory if we are more specific about what constitutes
risky behavior in social dilemmas. In contrast to previous research
on differences between public good and resource dilemmas, we do
not define risk in terms of the chances of creating the worst possible collective outcome (cf. Brewer & Kramer, 1986). Instead, we define risk in terms of variance, following other research on valence
effects (Kühberger, 1998), in which risky decision making involves
a choice between at least two options. The safe option has one or
738
P. de Heus et al. / Journal of Experimental Social Psychology 46 (2010) 736–742
more possible outcomes with values that are relatively close to
each other (low outcome variance), while the risky option has
two or more possible outcomes of which at least one is better than
the best outcome of the safe option, and at least one is worse than
the worst outcome of the safe option (high outcome variance).
Therefore, in the present article, a risky decision situation is defined
as one that demands a choice between options with different outcome variances. A risky choice is simply a choice for the high-variance option in such a risky decision situation.
In the following, we will argue that framing effects can be expected if and only if the option that might lead to the worst outcome is also the high-variance option. More specifically, we will
reason that in terms of high-variance of possible outcomes there
is no risky option in the prisoner’s dilemma, whereas there clearly
is such a high-risk option in the chicken game. As a consequence,
we expect to find clear and predictable valence effects in the chicken game, but not in the prisoner’s dilemma.
Current study
In our study, we use one-shot (i.e., games played only once) positively and negatively valenced versions of the prisoner’s dilemma
and the chicken game, for which the payoff structures are shown
in Fig. 1. What about risk in the prisoner’s dilemma? Inspection of
Fig. 1 shows that this prisoner’s dilemma is not a risky decision situation as defined above, because the two options (defect or cooperate) do not differ in terms of variance: whether one decides to
cooperate of defect, defection always yields higher outcomes than
cooperation. Of course, there is social uncertainty (i.e., uncertainty
about the decisions that others will make; Messick, Allison, &
Samuelson, 1988), because one has to decide without knowing
one’s opponent’s decision. Also, this uncertainty surely matters to
the players, because each player will always get two points more
if the other player chooses C instead of D. However, for each individual the relative attractiveness of C compared to D is independent of
the uncertain choice by the other player. Regardless of the other’s
decision, one will always earn one point more by choosing D instead
of C. In terms of outcome variance, there is no risky choice in the
sense of a choice between a relatively safe option and a more risky
option. Therefore, we see little reason to expect a valence effect.
The chicken game is a different story. Here defection is the highrisk (high-outcome variance) option, that might lead to the best,
but also to the worst possible payoff, whereas cooperation is relatively low-risk, because it can only bring the two intermediate payoffs. Therefore, taking risk in the sense of deciding whether or not
to gamble, is more essential in the chicken game than in the prisoner’s dilemma. As a consequence, we expect strong and predictable valence effects in the chicken game. Because in chicken
games defection is clearly the high-risk option, we expect more
defection in the loss frame than in the gain frame.
Our reasoning leads to three hypotheses. The first is basically a
manipulation check, but an important one. Participants in the
chicken game should see defection as a more risky choice than participants in the prisoner’s dilemma (Hypothesis 1). Our second
hypothesis is the central one, predicting a framing effect (more
cooperation in the gain frame than in the loss frame) in the chicken
game, but not in the prisoner’s dilemma (Hypothesis 2).
However, even if Hypotheses 1 and 2 are both confirmed, the possibility that a differential framing effect is caused by some other factor than perceived risk, has not been ruled out. One way of clarifying
the role of risk is to investigate the role of individual differences in
risk orientation (i.e., a general tendency to make risky instead of safe
choices; for recent discussions, see, for example, Meertens & Lion,
2008; Weber, Blais, & Betz, 2002). Little is known about the predictive power of risk orientation for cooperation in social dilemma
games (for exceptions, suggesting rather limited predictive power
in the iterated prisoner’s dilemma and the trust game respectively,
see Eckel & Wilson, 2004; van Assen & Snijders, 2004), but generally
we would expect that persons high on risk orientation more often
than persons low on risk orientation will choose the more risky option, if there is one. This leads to the expectation that risk orientation
predicts cooperative choice (more defection by high-risk seekers) in
the chicken game, but not in the prisoner’s dilemma (Hypothesis 3).
Finding such a game risk orientation interaction in addition to the
frame game interaction predicted by Hypothesis 2 would provide
additional support for the idea that the differential framing effects in
the prisoner’s dilemma and the chicken game are caused by the presence or absence of a risky option.
Method
Participants
Participants were 198 students from the University of Leiden, of
which 65 (33%) were male, with mean age 22.16 years (SD = 5.01).
Participants were invited to the laboratory for a study on decision
making. All participated voluntarily in our study (together with an
unrelated study that followed the present study) in exchange for
six euros. Data collection for the present study took about 20 min.
Design
The design of the study was a 2 2 game (prisoner’s dilemma
versus chicken game) frame (gain versus loss) factorial design,
with random assignment of participants to treatments, and choice
between cooperation and defection as the most important dependent variable.
Procedure
Upon arrival, participants were placed in separate cubicles with
a PC in it. After starting the computer program, the experimenter
explained that all communication with the experimenter and other
participants would be via the PC. To begin with, participants completed four questionnaires, of which only one, a measure of risk
orientation, is directly relevant to the present study. After completing these questionnaires, participants were informed that in the
next task they would play a kind of game with another participant
for lottery tickets, which gave a chance of winning a CD gift voucher worth 20 euros (about 25 US dollars). In order to make the
game as non-zero-sum as possible, it was pointed out that participant and other would not compete for the same prizes. It was also
explained that, to guarantee anonymity, all contact with the other
player would be via the computer, without face-to-face contact or
disclosure of identities afterwards.
Next, participants received instructions about the game (described below), followed by a single one-shot game (prisoner’s dilemma or chicken game) in which one had to choose between A
(cooperation) or B (defection); the words ‘‘cooperation” and
‘‘defection” were never used in the instructions. After this, without
being informed about the other’s choice, participants were asked a
few manipulation check questions about game comprehension and
framing, and a question about which choice (A or B) they perceive
as the most risky one. Finally, all participants were debriefed, paid,
and thanked for their participation.
Experimental manipulations
Framing manipulation
Our framing manipulation resembles in many respects the procedure used by Cropanzano, Paddock, Rupp, Bagger, & Baldwin
(2008). In the gain frame, participants were told that depending
739
P. de Heus et al. / Journal of Experimental Social Psychology 46 (2010) 736–742
Table 1
Hierarchical logistic regression analysis effects of frame, game, and risk orientation on cooperative choice.
Predictor step
Step 1
Frame
Game
Step 2
Frame game
Step 3
Risk orientation
Step 4
Game risk
Frame risk
Step 5
Game frame risk
**
Walda
B
p
Odds ratio
Model chi-square
13.57
.58
.98
3.52
10.01
.061
.002
1.78
.38
1.46
4.89
.027
.23
.62
6.48
.011
.54
1.45
.80
6.40
2.35
.011
.125
4.25
.45
.35
.08
.785
1.42
**
Nagelkerke R square
.091
18.73***
.124
25.63***
.166
33.93***
.216
34.00***
.216
p < .01.
p < .001.
a
Wald df are always 1. Model df are 2, 3, 4, 6 and 7 for steps 1–5, respectively.
***
on choices by self and other, they could win zero, one, two, or three
lottery tickets. In the loss frame, they were told that they would
start with three lottery tickets, but that depending on their choices,
they would lose between zero and three tickets. The exact numbers
of points (tickets) gained or lost for both games are depicted in
Fig. 1. In addition to this general instruction, the prisoner’s dilemma and the chicken game were consistently described in terms of
winning (gain frame) or losing tickets (loss frame).
Game manipulation
The instructions for both games consisted of a payoff matrix
that was permanently visible on screen, and a verbal explanation
of the game. In the description for both games, the instructions
from the gain frame will be presented, with alternative formulations from the loss frame between brackets. Both games started
as follows.
‘‘In the next part, you will play for points, and each point represents a lottery ticket. How many points you will win [lose], will be
determined not only by your own choice, but also by the choice of
the other person. At the same time, his or her points are also partly
determined by your choices. The game will be played as follows.
You will have to make a choice between two possibilities, called
A and B (the other will have to make the same choice). At the moment you make your choice, you do not know the choice of the
other, and the other does not know your choice. The number of
points that each of you wins [loses] is determined by the combination of choices by you and the other. As you can see in the figure
below, there are four possible outcomes.” Next, one of the four payoff matrices from Fig. 1 was shown (with A and B instead of C and
D), which remained on screen until participants had played their
game. In addition, outcomes for the four possible combinations
of choices were also explained in words. To make sure that participants understood the games, an explanation about the strategic
implications of the payoff structure was given to the participant.
In the prisoner’s dilemma conditions it was explained that choosing
B always leads to a higher gain [smaller loss] for oneself than
choosing A (regardless of the other’s choice), but that if both players choose B, both will be worse off than if both had chosen A. Participants in the chicken game were told that choosing B might lead
to the best possible outcome (if the other chooses A), but also to
the worst possible outcome (if the other chooses B) for oneself,
whereas choosing A can only lead to the second-best and secondworst outcomes.
Risk orientation
As a measure of risk orientation, we administered the Risk Orientation Questionnaire (ROQ: Rohrmann, 2002). The ROQ contains
twelve domain-independent items (e.g. ‘‘Even when I know that
my chances are limited, I try my luck”), each with a seven-point
answering scale from 1 = ‘‘does not at all apply to me” to 7 = ‘‘very
much applies to me”. The ROQ consists of two moderately negatively correlated (correlations around .35) subscales for Risk Propensity (seeking risks) and Cautiousness (avoiding risks). Our final
measure of risk orientation was the average of all 12 items (with
reversed scoring of all Cautiousness items), so higher scores indicated a higher willingness to take risks.
Dependent measures
The two dependent measures were cooperative choice (did one
choose cooperation or defection in the one-shot game?) and most
risky option (did one indicate that cooperation (A) or defection
(B) was the more risky option in this game?).
Results
Manipulation checks
According to chi-square tests, there were no significant differences in understanding between the prisoner’s dilemma (PD) and
the chicken game (CG). Most participants correctly indicated in
both games that one-sided defection brought the best possible outcome for self (78.0% and 80.6% in the PD and CG respectively), that
one-sided cooperation in the PD and mutual defection in the CG led
to the worst outcome for self (83.0% and 76.5%), and that in both
games mutual cooperation brought the best joint outcome for self
and other (87.0% and 93.9%), so generally game understanding appeared to be adequate. The framing manipulation was checked by
comparing the gain and loss conditions on the question whether
they felt they could gain or lose something by playing the game.
Participants in the gain frame condition predominantly reported
that they could gain something (87.9%), whereas participants in
the loss frame condition predominantly reported that they could
lose something (67.7%). These findings show that the manipulations were perceived as intended.1
Most risky option
To check if defection was recognized as the more risky option in
the CG, but not in the PD, a chi-square test (with continuity correc1
The fact that a minority of the participants in the loss condition did report that
they could gain something may reflect that participants also compared outcomes to
how they entered the lab (i.e., they could never end up with less money than they
possessed when they entered the laboratory).
740
P. de Heus et al. / Journal of Experimental Social Psychology 46 (2010) 736–742
tion) was performed on the 2 2 cross table of game (PD versus
CG) by the question about the most risky option (cooperation versus defection). The chi-square test was highly significant,
X2(1) = 76.60, p < .001, indicating that defection was much more
often seen as the most risky choice in the CG (87.8%) than in the
PD (25.0%), so Hypothesis 1 was confirmed.
The main effect of risk orientation in step 3 was significant,
B = .62, Wald(1) = 6.48, p < .05, indicating that a higher willingness
to take risks leads to less cooperative choices. This main effect was
moderated by a significant game risk orientation interaction (step
4), B = 1.45, Wald(1) = 6.40, p < .05, which demonstrated that the effect of risk orientation on cooperation was different in the PD and the
CG. Separate logistic regression analyses for the two games, with
frame and risk orientation as predictors and cooperative choice as
the dependent variable, revealed that a higher risk orientation led
to less cooperative choice in the CG, B = 1.48, Wald(1) = 9.74,
p < .01, but not in the PD, B = .20, Wald(1) = .45, p > .50. An alternative, more ANOVA-like way of describing the interaction is that in the
CG defectors were higher in risk orientation than cooperators
(M = 4.90 versus 4.32), while there was no such difference in the
PD (M = 4.41 versus 4.32). The frame risk orientation (step 4)
and the threeway (step 5) interactions were nonsignificant (Table 1).
These results are consistent with Hypothesis 3.
Cooperative choice
Discussion
To test our central prediction of a framing effect (more cooperation in gain frame than in loss frame) in the chicken game, but not
in the prisoner’s dilemma, a hierarchical logistic regression analysis was performed, with game and frame as independent variables,
and cooperative choice as the dependent variable. The analysis was
performed hierarchically, because in logistic regression analysis,
regression weights and significance tests of predictors can not be
interpreted as main effects when their products (interaction terms)
are also predictors in the analysis (Jaccard, 2001). Because interactions in logistic regression analysis might lead to inconsistencies
between interpretation of proportions (which fit with how humans
tend to think about effects) and interpretation of logits (i.e., the
natural logarithms of the odds for those proportions, which fit with
the estimated parameters in logistic regression; Ganzach, Saporta,
& Weber, 2000), the amount of cooperation is given both in proportions and in logits of cooperative choice. The results are presented
in Table 1 (regression weights, odds ratios, and significance tests)
and Table 2 (proportions and logits).
In step 1, only the main effect for game was significant,
B = .988, Wald(1) = 10.01, p < .01, indicating that cooperation
was more often chosen in the chicken game (74.5%) than in the
prisoner’s dilemma (53.0%). In step 2, the frame game interaction
proved to be significant, B = 1.46, Wald(1) = 4.89, p < .05. Inspection of cell proportions (Table 2) showed that while there was a
strong framing effect in the chicken game (87.5% cooperative
choice in the gain frame versus 62.0% in the loss frame), there
was no framing effect in the prisoner’s dilemma (52.9% versus
53.1%), with logits showing the same pattern as proportions
(Table 2). Separate chi-square tests (with continuity correction)
for the relationship between frame and choice in the two games,
showed that the difference between gain and loss frames was
highly significant in the chicken game, X2(1) = 7.09, p < .01, but absent in the prisoner’s dilemma, X2(1) = .00, p > .99. These results are
in agreement with Hypothesis 2.
Are cooperation and defection related to valence? Our comparison between the prisoner’s dilemma and the chicken game was
supportive for our idea that to find valence effects in social dilemmas, the dilemma must offer a choice between a relatively sure
and a relatively risky option. Participants identified defection as
the more risky option in the chicken game, but not in the prisoner’s
dilemma (Hypothesis 1). Moreover, this difference between the
two dilemmas also was related to choice behavior, as a higher willingness to take risks led to more defection in the chicken game, but
not in the prisoner’s dilemma (Hypothesis 3). Most importantly, a
valence effect (more cooperation in the gain frame than in the loss
frame) was found in the chicken game, but not in the prisoner’s dilemma (Hypothesis 2).
Taken together, our findings suggest that prospect theory can in
fact be meaningfully applied to social dilemmas (cf. van Dijk &
Wilke, 1995). The prerequisite for this appears to be that the
dilemma should clearly distinguish between more risky and less
risky options. This is true in the chicken game, but not in the
prisoner’s dilemma. If our reasoning about differential framing
effects in prisoner’s dilemma and chicken game is correct, could
this explain the inconsistent results of framing in social dilemmas
studies? Although we do not think it is the only explanation, it
might help. To see how, let us return to the argument that in public
good dilemmas and resource dilemmas, cooperation is relatively
safe, whereas defection is relatively risky (Brewer & Kramer,
1986). The underlying rationale for this assumption appears to
be that the worst thing that can happen is that the public good is
not realized or that the resource becomes depleted. Defection
(i.e., giving little or taking much) is the risky choice, because it
makes this worst possibility more likely. If individuals are riskseeking in the loss frame and risk-averse in the gain frame, they
should be more inclined toward defection in a public good game
(loss frame) than in an equivalent commons dilemma (gain-frame).
However, as we reasoned before, in order to apply prospect theory to social dilemmas, it should be clear what constitutes the risky
option and what constitutes the safe(r) option. Many social dilemmas are like N-person prisoner’s dilemmas, in which it is ambiguous
which is the more risky option: choosing defection with a larger
chance that the public good remains unrealized, or choosing cooperation with a (somewhat smaller, but still substantial) chance that
the public good remains unrealized despite one’s own sacrifices
(for which one receives no compensation whatsoever). Perhaps
we can not expect framing effects in social dilemmas, unless they
are like N-person chicken games, in which the worst outcome is
for a defector if the public good remains unrealized.
Table 2
Proportions and logits of cooperative choices for different games and frames.
Dependent measure
Frame
PD
CG
Total
Proportions
Gain
Loss
Total
.529
.531
.530
.875
.620
.745
.697
.576
.636
Logits
Gain
Loss
Total
.116
.124
.120
1.946
.490
1.072
.833
.305
.560
Risk orientation
To test Hypothesis 3 (higher risk orientation leads to more
defection in the chicken game, but not in the prisoner’s dilemma),
the hierarchical logistic regression analysis described above was
extended with three more steps. In step 3, risk orientation was
added as a predictor, followed in step 4 by the game risk orientation and frame risk orientation interactions, and in step 5 by
the threeway interaction (Table 1).
P. de Heus et al. / Journal of Experimental Social Psychology 46 (2010) 736–742
At this point it is also useful to discuss some limitations of the
present study. Whereas the results support our idea that reliable
valence effects can be observed in social dilemmas when choice
options differ in terms of outcome variance, it may of course be relevant to see whether our results will be replicated in future studies. If our reasoning about games with and without a risky option is
correct, we would expect our results to generalize beyond the prisoner’s dilemma and the chicken game to other experimental
games, of which four deserve special mention. The first is generalization to prisoner’s dilemmas and chicken games with continuous
choices (i.e., in which different degrees of cooperation can be chosen). Because binary dependent variables as in the present study
are usually bad for statistical power, we would expect stronger effects in such continuous games. The second generalization might
be from one-shot to iterated prisoner’s dilemmas and chicken
games against the same opponent. However, because with iterated
games the numbers of possible strategies and outcomes grow very
fast with the number of repetitions, it remains to be tested
whether the distinction between choosing options with high versus low outcome variance (chicken game) versus choosing options
with equal outcome variance (prisoner’s dilemma) will generalize
to repeated games. The third generalization is to another two-person game, the trust game (Dasgupta, 1988), an asymmetric game in
which one player (the trustor) has to choose between a risky option (trust) and a completely sure thing (no trust), providing an
even sharper test than the present study for our hypothesis that
it is the presence or absence of a (relatively) sure thing versus variation in possible outcomes that predicts the presence or absence
of framing effects. The fourth generalization is to larger groups,
from two-person to N-person prisoner’s dilemmas and chicken
games. Finding a framing effect in N-person chicken, but not in
the N-person prisoner’s dilemma would provide additional empirical support for our explanation of inconsistent results in the literature on framing in social dilemmas. In such studies, it may also be
worthwhile to obtain additional evidence for the assumed underlying process, e.g. by measuring the perceived variance of choice options. One should realize, however, that such measurements may
neither be necessary (since people may react to circumstances
without being able to describe them; e.g. Nisbett & Ross, 1980),
nor be easy, as they require participants to ‘‘reproduce” what
may be considered a rather abstract concept.
In the current paper, we obtained additional evidence for our
reasoning by measuring the participants’ risk orientation with
the ROQ. There are many different measures for risk orientation,
some involving choices between sure things and gambles or between different kinds of gambles (e.g. van Assen & Snijders,
2004), some very domain-specific (e.g. Weber et al., 2002), and
some, like the ROQ (Rohrmann, 2002) measuring a general, domain-independent attitude toward taking risks. Future research
will have to show whether other measures of risk orientation show
the same relationships with cooperation in social dilemmas. In
such future research, we may also investigate whether risk orientation correlates with other relevant constructs, and whether such
relations might (partly) explain our findings. In this context, it may
be relevant that the willingness to take risks has been related to social value orientations, i.e., to the relative weights that people put
on their own versus other’s outcomes. For example, Or-Chen and
Suleiman (2003) reported that prosocials (i.e., those inclined to
cooperate) are more risk-averse. Comparing this to our finding that
– only in the chicken game – a high-risk orientation was associated
with less cooperation, might make one wonder whether this finding reflects a differential concern for other’s outcomes. However,
the absence of any effect of risk orientation in the prisoner’s dilemma does not match with this idea (especially since social value orientations have produced strong and reliable effects in social
dilemmas; see van Lange, De Cremer, van Dijk, & van Vugt,
741
2007). To be sure, we also measured social value orientations in
our study, and did not find any relationship with risk orientation
or cooperation (see also Kanagaretnam, Mestelman, Nainar, &
Shehata, 2009). Nevertheless, for future research it may be useful
to investigate the connection between risk orientation and other
constructs (e.g., trust, sensation seeking). Such studies may further
increase our understanding of how prospect theory can be meaningfully applied to the field of social dilemmas.
References
Axelrod, R. (1984). The evolution of cooperation. New York: Basic Books.
Brewer, M. B., & Kramer, R. M. (1986). Choice behavior in social dilemmas: Effect of
social identity, group size, and decision framing. Journal of Personality and Social
Psychology, 50, 543–549.
Cropanzano, R., Paddock, L., Rupp, D. E., Bagger, J., & Baldwin, A. (2008). How
regulatory focus impacts the process-by-outcome interaction for perceived
fairness and emotions. Organizational Behavior and Human Decision Processes,
105, 36–51.
Dasgupta, P. (1988). Trust as commodity. In D. Gambetta (Ed.), Trust: Making and
breaking cooperative relations (pp. 49–72). Oxford: Blackwell.
Eckel, C. C., & Wilson, R. K. (2004). Is trust a risky decision? Journal of Economic
Behavior & Organization, 55, 447–465.
Ganzach, Y., Saporta, I., & Weber, Y. (2000). Interaction in linear versus logistic
models: A substantive illustration using the relationship between motivation,
ability, and performance. Organizational Research Methods, 3, 237–253.
Jaccard, J. (2001). Interaction effects in logistic regression. Thousand Oaks (CA): Sage.
Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under
risk. Econometrica, 47, 363–391.
Kahneman, D., & Tversky, A. (1984). Choices, values, and frames. American
Psychologist, 39, 341–350.
Kanagaretnam, K., Mestelman, S., Nainar, K., & Shehata, M. (2009). The impact of
social value orientation and risk attitudes on trust and reciprocity. Journal of
Economic Psychology, 30, 368–380.
Komorita, S. S., & Carnevale, P. J. (1992). Motivational arousal vs. decision framing in
social dilemmas. In W. B. G. Liebrand, D. M. Messick, & H. A. M. Wilke (Eds.),
Social dilemmas (pp. 209–223). New York: Pergamon.
Komorita, S. S., & Parks, C. D. (1995). Interpersonal relations: Mixed motive
interactions. Annual Review of Psychology, 46, 183–207.
Kopelman, S., Weber, J. M., & Messick, D. M. (2002). Factors influencing cooperation
in commons dilemmas: A review of experimental psychological research. In E.
Ostrom, T. Dietz, N. Dolsak, P. C. Stern, S. Stonich, & E. U. Weber (Eds.), The
drama of the commons (pp. 113–156). Washington, DC: National Academy Press.
Kühberger, A. (1998). The influence of framing on risky decisions: A meta-analysis.
Organizational Behavior and Human Decision Processes, 75, 23–55.
Liebrand, W. B. G., Wilke, H. A. M., Vogel, R., & Wolters, F. J. M. (1986). Value
orientation and conformity: A study using three types of social dilemma games.
Journal of Conflict Resolution, 30, 77–97.
Maynard Smith, J., & Price, G. R. (1973). The logic of animal conflict. Nature, 246,
15–18.
McCusker, C., & Carnevale, P. J. (1995). Framing in resource dilemmas: Loss aversion
and the moderating effect of sanctions. Organizational behavior and Human
Decision Processes, 61, 190–201.
Meertens, R. M., & Lion, R. (2008). Measuring an individuals tendency to take risks:
The risk propensity scale. Journal of Applied Social Psychology, 38, 1506–1520.
Messick, D. M., Allison, S. T., & Samuelson, C. D. (1988). Framing and communication
effects on group members’ responses to environmental and social uncertainty.
In S. Maital (Ed.). Applied behavioral economics (Vol. 2, pp. 677–700). New York:
New York University Press.
Messick, D. M., & Brewer, M. B. (1983). Solving social dilemmas: A review. In L.
Wheeler & P. Shaver (Eds.). Review of Personality and Social Psychology (Vol. 4,
pp. 11–44). Beverly Hills, CA: Sage.
Murnighan, J. K. (1991). The dynamics of bargaining games. Englewood Cliffs, NJ:
Prentice Hall.
Nisbett, R. E., & Ross, L. (1980). Human inference: Strategies and shortcomings of social
judgment. Englewood Cliffs, NJ: Prentice-Hall.
Or-Chen, K., & Suleiman, R. (2003). Pro-socials are more risk-averse than others.
When? And why? In Paper presented at the 10th international conference on social
dilemmas, Marstrand, Sweden.
Rapoport, A., & Chammah, A. M. (1969). The game of chicken. In I. R. Buchler & H. G.
Nuttini (Eds.), Game theory in the behavioral sciences (pp. 151–175). Pittsburgh:
University of Pittsburgh Press.
Rohrmann, B. (2002). Risk attitudes scales: Concepts and questionnaires. Project
Report, University of Melbourne/Australia. Retrieved June 20, 2008, from
University of Melbourne, Department of Psychology. Web site: http://
www.rohrmannresearch.net/pdfs/rohrmann-ras-report.pdf.
Rutte, C. G., Wilke, H. A. M., & Messick, D. M. (1987). The effect of framing social
dilemmas as give-some or take-some games. British Journal of Social Psychology,
26, 103–108.
van Assen, M., & Snijders, C. (2004). Effects of risk preference in social dilemmas. A
game-theoretical analysis and evidence from two experiments. In R. Suleiman,
D. V. Budescu, I. Fischer, & D. M. Messick (Eds.), Contemporary psychological
742
P. de Heus et al. / Journal of Experimental Social Psychology 46 (2010) 736–742
research on social dilemmas (pp. 24–57). Cambridge: Cambridge University
Press.
van Dijk, E., & Wilke, H. A. M. (1995). Coordination rules in asymmetric social
dilemmas: A comparison between public good dilemmas and resource
dilemmas. Journal of Experimental Social Psychology, 31, 1–27.
van Lange, P. A. M., De Cremer, D., van Dijk, E., & van Vugt, M. (2007). Self-interest
and beyond: Basic principles of social interaction. In A. W. Kruglanski & E. T.
Higgins (Eds.), Social psychology: Handbook of basic principles (2nd ed.,
pp. 540–561). New York: Guilford.
Weber, E. U., Blais, A. R., & Betz, N. E. (2002). A domain-specific risk-attitude scale:
Measuring risk perceptions and risk behaviors. Journal of Behavioral Decision
Making, 15, 263–290.
Wit, A. P., & Wilke, H. A. M. (1992). The effect of social categorization on cooperation
in three types of social dilemmas. Journal of Economic Psychology, 13, 135–151.